Bet Hedging in Yeast by Heterogeneous, Age-CorrelatedExpression of a Stress ProtectantSasha F. Levy*¤, Naomi Ziv, Mark L. Siegal*
Center for Genomics and Systems Biology, Department of Biology, New York University, New York, New York, United States of America
Abstract
Genetically identical cells grown in the same culture display striking cell-to-cell heterogeneity in gene expression and othertraits. A crucial challenge is to understand how much of this heterogeneity reflects the noise tolerance of a robust systemand how much serves a biological function. In bacteria, stochastic gene expression results in cell-to-cell heterogeneity thatmight serve as a bet-hedging mechanism, allowing a few cells to survive through an antimicrobial treatment while othersperish. Despite its clinical importance, the molecular mechanisms underlying bet hedging remain unclear. Here, weinvestigate the mechanisms of bet hedging in Saccharomyces cerevisiae using a new high-throughput microscopy assay thatmonitors variable protein expression, morphology, growth rate, and survival outcomes of tens of thousands of yeastmicrocolonies simultaneously. We find that clonal populations display broad distributions of growth rates and that slowgrowth predicts resistance to heat killing in a probabalistic manner. We identify several gene products that are likely to playa role in bet hedging and confirm that Tsl1, a trehalose-synthesis regulator, is an important component of this resistance.Tsl1 abundance correlates with growth rate and replicative age and predicts survival. Our results suggest that yeast bethedging results from multiple epigenetic growth states determined by a combination of stochastic and deterministicfactors.
Citation: Levy SF, Ziv N, Siegal ML (2012) Bet Hedging in Yeast by Heterogeneous, Age-Correlated Expression of a Stress Protectant. PLoS Biol 10(5): e1001325.doi:10.1371/journal.pbio.1001325
Academic Editor: Laurence D. Hurst, University of Bath, United Kingdom
Received September 16, 2011; Accepted March 26, 2012; Published May 8, 2012
Copyright: � 2012 Levy et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricteduse, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: SFL was supported by NIH Ruth Kirchstein individual postdoctoral fellowship F32GM79909 and American Cancer Society postdoctoral fellowship PF-10-028-01-CCG. This work was funded by NSF grant IOS-0642999 to MLS. The funders had no role in study design, data collection and analysis, decision to publish, orpreparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
Abbreviations: AER, age expression ratio; GFP, green fluorescent protein; PFS, perfect focus system; SD, standard deviation; SEM, standard error of the mean;WGA-TRITC, wheat-germ agglutinin tetramethyl rhodamine isothiocyanate
* E-mail: [email protected] (SFL); [email protected] (MLS)
¤ Current address: Department of Genetics, Stanford University Medical School, Stanford, California, United States of America
Introduction
Clonal populations of cells grown in a constant environment
display a striking amount of cell-to-cell heterogeneity. For
example, in bacteria, yeast, and mammalian cell lines, levels of
some gene products vary widely between cells [1–5]. A crucial
challenge is to understand how much of this heterogeneity serves a
biological function [6,7]. That is, does variability in gene
expression between clonal cells simply reflect the noise tolerance
of a robust system, or does the variation itself increase population
fitness?
In bacteria, several examples exist in which clonal variation in gene
expression correlates with a morphological or physiological state that
presumably confers a fitness advantage in some environments. These
examples include competence to uptake foreign DNA [8–10],
initiation of sporulation [11,12], and expression of cell surface pili
[13,14]. In each case, a binary fate decision is controlled in part by
stochastic expression of a crucial regulatory protein.
A population-fitness advantage for heterogeneity is even more
obvious for the phenomenon known as bacterial persistence. When
a clonal population of Escherichia coli is exposed to a lethal dose of
ampicillin, the vast majority of the population dies at a fast
exponential-decay rate but rare, slow-growing ‘‘persister’’ cells die
at a much slower rate [15,16]. These persister cells can subsequently
switch to the common, fast-dividing state, thereby restoring the
population after removal of the antibiotic. Persistence is therefore
considered a canonical example of a bet-hedging mechanism (Box
1), whereby a population maximizes its long-term fitness in an
unpredictably changing environment by distributing risk among
individuals [16,17]. In a benign environment, most E. coli cells adopt
the sensible strategy of fast growth, whereas a small proportion of
cells adopt the high-risk strategy of entering the persister state,
which could reap large benefits should the environment change.
Single-cell observations in a microfluidic chamber suggest that,
as with competence and the other examples above, persisters and
non-persisters constitute binary states that interconvert through a
stochastic mechanism [16]. However, despite the clinical impor-
tance of persistence, and despite indications that slow growth
might be a general means of surviving stress [18–20], the
molecular mechanisms underlying persistence remain unclear
[21,22]. This is due, in large part, to the experimental difficulty of
identifying and characterizing a rare persister subpopulation prior
to antimicrobial treatment.
Like clonal populations of bacteria, those of the budding yeast S.
cerevisiae have also been shown to contain a large amount of cell-to-
cell heterogeneity [3,4,23,24]. In both bacteria and yeast, one
PLoS Biology | www.plosbiology.org 1 May 2012 | Volume 10 | Issue 5 | e1001325
component of gene-expression variation is so-called ‘‘intrinsic’’
noise, which is operationally defined as fluctuations that are not
correlated between identical promoters in the same cell [3]. In
yeast, as in other eukaryotes, an important component of intrinsic
noise is fluctuation between more or less accessible chromatin
states [3,5,25–28]. Mutations in yeast genes associated with
chromatin remodeling alter the extent of heterogeneity in both
protein expression [3,29] and cell morphology [30].
By contrast, ‘‘extrinsic’’ noise is defined as variation that is
correlated between different alleles of the same gene, or between
different genes [3]. Such variation reflects either fluctuations in the
concentrations of upstream regulators (i.e., intrinsic noise
upstream can produce extrinsic noise downstream), or fluctuations
in global cell state, such as the abundances of ribosomes [31] or
mitochondria [32].
In yeast, evidence suggests that a fraction of what might
operationally be defined as extrinsic noise is instead due to
deterministic factors. For example, fluctuations of many gene
products have been found to correlate with the cell cycle [4] and
cell size [3,33,34]. Additionally, unequal segregation of certain
molecular components between mother and daughter cells [35–
38] or daughter-specific expression [39] could produce meaningful
replicative-age-dependent heterogeneity within a yeast population
[24]. For example, cells that have undergone ,eight replicative
cycles survive ultraviolet irradiation better than younger or older
cells [40].
A combination of stochastic and deterministic influences could
provide the basis for more complex bet-hedging mechanisms than
the binary switches that appear to be primary in bacteria. Indeed,
the pathogenic yeast Candida albicans displays at least seven
different metastable colony morphologies when grown on agar
[41]. Another opportunistic pathogen, C. glabrata, which despite its
name is actually a member of the Saccharomyces clade, shows similar
multi-stability [42,43]. It should also be kept in mind that bacterial
bet-hedging mechanisms might be more complex as well, and that
the apparent primacy of binary switches might be a product of the
phenotypes chosen for study and of experimental limitations in
phenotypic measurement. For example, although E. coli antibiotic
persistence is commonly described as a two-state system, recent
observations of macroscopic bacterial colonies on agar have found
a continuous distribution of growth rates [44]. Additionally,
asymmetric cell division has been found to underlie bet hedging to
starvation in the bacterium Sinorhizobium meliloti, indicating that
deterministic factors may be important in prokaryotes as well [45].
Here, we investigate the mechanisms of bet hedging and
persistence in S. cerevisiae using a new high-throughput microscopy
assay capable of monitoring variable protein expression, morphol-
ogy, growth rate, and survival outcomes of tens of thousands of
yeast microcolonies simultaneously. We find that clonal popula-
tions of yeast grown in a rich, benign environment display a wide
and continuous distribution of growth rates that can be modulated
by mutations in genes involved in chromosome organization or
other core regulatory functions. Using a bioinformatic screen, we
identify candidate gene products whose expression correlates with
growth rate and establish that Tsl1, a protein involved in the
synthesis of the disaccharide trehalose, is a molecular marker for
slow growth in the benign environment. Using quantitative
measurements of microcolony growth rates and abundance of
fluorescently tagged Tsl1, we show that both slow growth and Tsl1
abundance predict survival of heat stress in a graded rather than
binary fashion and that Tsl1 is an important component of the
stress survival. Lastly, we investigate replicative age as a potential
source of heterogeneity in this stress-survival system and in protein
expression in general. We find that Tsl1-abundant cells tend to be
older and, more generally, that replicative age is an underlying
component of cell-to-cell variation in the expression of many
proteins.
Results
High-Throughput Microcolony Growth AssayMicrobial fitness assays have historically been limited to
ensemble measurements that calculate the difference in mean
growth rate or the competitive fitness advantage of one population
over another. Besides suffering severe limitations in the number of
replications that are experimentally feasible, these assays do not
measure the variance of growth rates, even though this is likely to
be an evolutionarily meaningful parameter in both static and
fluctuating environments and over the course of population
bottlenecks [17,24,46–49].
To overcome these limitations, we developed a high-throughput
assay that measures microcolony growth by time-lapse bright-field
microscopy (Figure 1A; Videos S1 and S2). Exponentially growing
cells are plated at a low density in rich, liquid medium on glass-
bottomed micro-well plates and allowed to grow into isolated
microcolonies of up to ,100 cells (Materials and Methods).
During this growth period, 1-h time-lapse images of ,3,000 low-
magnification fields are captured in parallel allowing for simulta-
neous observation of ,105 microcolonies. Custom-written image
analysis software tracks changes in area over time, and these
measurements are used to calculate the specific growth rate of
each microcolony (the change in the log of the area per hour).
Each growing microcolony displays log-linear growth over the
period of observation (Materials and Methods), yet different
microcolonies grow at vastly different rates (Figure 1B and 1C).
The automated measurements of microcolony area correlate
extremely well with manual cell counts over a range of growth
rates (R2.0.9) (Figures 1C, S1, and S2), indicating that changes in
area are representative of cell-division rates. Growth-rate distri-
butions generated from all individual microcolony growth rates
measured within a well of a 96-well plate are highly reproducible
Author Summary
Genetically identical cells grown in the same environmentcan display heterogeneity in their morphology, behavior,and composition of their cellular components. In somemicroorganisms, such cellular heterogeneity can underlie aphenomenon known as bet hedging because it enablessome cells to survive in harsh environments, henceincreasing the overall population fitness when environ-mental shifts are unpredictable. Bet hedging is likely to bean important strategy by which microbes infect humansand evade antimicrobial treatments, yet little is known ofhow cellular heterogeneity contributes to microbial sur-vival. Here, we study the mechanisms underlying bethedging in yeast. We find that populations of geneticallyidentical yeast contain a broad distribution of growth ratesand that slow growth predicts resistance to heat killing in agraded fashion. We identify several gene products that arelikely to play a role in this bet-hedging strategy andconfirm that Tsl1, a regulator of the production of thedisaccharide trehalose, is an important component ofacute stress resistance. Finally, we find that old age in cellscorrelates with a Tsl1-abundant, stress-resistant cell state.Our results suggest that trehalose synthesis is part of acomplex and multifactorial mechanism that underlies bethedging in yeast.
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 2 May 2012 | Volume 10 | Issue 5 | e1001325
between wells on a single plate or between experimental days
(Figure S3).
In wild-type populations grown in a benign environment, a
large fraction of microcolonies grow at less than half the median
population growth rate (1.3%–10%, depending on the strain)
(Figure 1D; Table S1). Because growth rate is extremely consistent
within a microcolony over the duration of tracking, this wide
distribution indicates that substantial differences in growth rate
between isogenic cells exist and are heritable over several
generations. We hypothesized that, as in bacteria, cell growth
rates constitute a phenotypically observable component of
epigenetic cell states that together act as a bet-hedging mechanism
in yeast. That is, the lowered relative fitness of slow-growing cells
in the benign environment would have an increased relative fitness
in other, perhaps harsher, environments, allowing a clonal
population to maximize the population fitness over multiple
environments. We first ruled out several alternative technical and
biological explanations of slow growth. One possibility is that local
nutrient depletion by neighboring microcolonies causes closely
spaced microcolonies to grow slower than distantly spaced
microcolonies. With the exception of microcolonies within
35 mm (4–8 cell lengths) of each other, microcolony growth rate
distributions showed no observable dependence on the proximity
of a microcolony to its nearest neighbor (Figure S4). A slight
difference in growth-rate distribution of microcolonies that fall
within 35 mm of each other could be detected and is likely due to a
technical bias of the experiment rather than local nutrient
depletion (Materials and Methods). Regardless of the cause,
removing closely spaced microcolonies had a minimal effect on
observed growth rate distributions. Nonetheless, to be conserva-
tive, we ignored these microcolonies in all data reported here.
A second possible explanation for the frequent occurrence of
slow-growing microcolonies is that these cells are petites, having
lost mitochondrial function. Such losses can occur frequently in
yeast [50]. To test this possibility, we generated growth rate
distributions of single-deletion strains of several genes necessary for
Box 1. Bet Hedging: Definitions and Open Questions
Bet hedging is an often loosely used term to describe a risk-spreading strategy that increases a population’s fitness inunpredictably fluctuating environments. A rigorous definitionof bet hedging is reversible epigenetic phenotypic heteroge-neity that results in decreased arithmetic mean fitness andincreased geometric mean fitness of the population acrossenvironments [87]. This concept can be best understood inmicrobes that compete in a benign environment most of thetime but unpredictably encounter a harsh environment. In thecommon benign environment, a heterogeneous populationwill be less fit than one with a single robust phenotype tunedto the benign environment. However, when acute transitionsto the harsh environment occur, the heterogeneous popula-tion will contain some individuals better able to cope with it,and thereby will outcompete the robust population.
The requirements for an experimental demonstration of bethedging are currently undergoing a lively debate [88]. Generalagreement underlies several criteria: (1) bet hedging isepigenetic in nature and therefore must be demonstrated inisogenic lines; (2) bet hedging must be demonstrated to actacross unpredictable environmental shifts where it could bereasonably assumed that a sense-and-response system wouldhave greater costs than benefits [17]; (3) cell lineages must bedemonstrated to interconvert between phenotypic statesreversibly and independently of the prevailing environment;and (4) alternative phenotypic states must be demonstratedto confer different fitnesses across environments.
Several other proposed requirements for bet hedging lackconsensus. The first concerns whether or not interconvertingphenotypic states must be binary. Early examples of putativebet-hedging systems in bacteria meet this criterion [8–16].For example, a discrete slow-growth phenotype in E. colipredicts survival of high doses of ampicillin [15,16]. However,several examples in yeast demonstrate multiple metastablephenotypic states [41–43], and recent observations ofmacroscopic bacterial colonies on agar have found acontinuous distribution of growth rates [44]. Heretofore, itremained an open question whether multiple discretephenotypes or a continuous distribution of phenotypescould act as a risk-spreading mechanism.
A second debate surrounds the mechanism by which cellsinterconvert between phenotypic states. Because phenotyp-
ic switching must be insensitive to the environment, it hasbeen generally assumed that the mechanism underlyingswitching must be stochastic [16,17,89]. We argue here thatdeterministic factors, such as replicative age, could alsounderlie bet hedging. For example, unequal segregation ofcertain molecular components between mother and daugh-ter cells [35–38] or daughter-specific expression [39] couldproduce meaningful replicative-age-dependent fitness dif-ferences within a yeast population that is independent ofenvironmental shifts [24]. Indeed, a completely deterministicasymmetric cell division has been found to underlie fitnessdifferences under starvation in the bacterium S. meliloti [45].
A third debate surrounds the selective forces that ultimatelyproduce a bet-hedging system. That is, if a distribution ofphenotypes and relative fitnesses is produced as a by-productof some other adaptive event, can this be classified as a bet-hedging system? Or, must selection be for the distributedphenotypes per se? For example, old cells might be selectedto increase production of chaperones to compensate for anincreased misfolded-protein load. A by-product of theincreased abundance of chaperones might be an increasedheat tolerance of old cells and the observation of survivalheterogeneity within a population exposed to acute heatstress. To demonstrate conclusively that survival heterogene-ity is a consequence of selection for distributed phenotypes,however, the environmental regime and fitness distributionsmust be measured during the adaptation itself [45]. Thisrepresents an extreme experimental challenge that has beenovercome only a few times [89–91]. Indeed, in a comprehen-sive survey of over 100 studies of candidate bet-hedgingsystems, Simons [92] found only four cases meeting his moststringent evidence criteria. We therefore propose to reservethe term ‘‘adaptive bet hedging’’ [93,94] for cases wheresurvival heterogeneity has been demonstrated to be conse-quence of selection for distributed phenotypes and use theterm ‘‘bet hedging’’ in all other cases where the consensuscriteria have been met. In the present study, we showreversible and environment-independent interconversion ofphenotypic states that results in heterogeneity of survival forisogenic cells exposed to acute stress. We therefore refer tothis phenomenon as bet hedging, while acknowledging thatthe hypothesis of its being adaptive remains to be tested.
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 3 May 2012 | Volume 10 | Issue 5 | e1001325
Figure 1. A new high-throughput assay to measure growth rate variance in yeast. (A) A schematic of the assay. Cells in logarithmic growth areplated at a low density on a glass-bottomed multi-well plate. Low-magnification time-lapse bright-field images are captured in a highly parallelized manner.Custom-written software tracks colony area over time. (B) Isogenic cells grow at different rates. Time-lapse images of a portion of one field (left) and theoutput of image analysis software (right). (C) Microcolony area correlates with cell number. Representative traces of fast- (purple), medium- (grey), and slow-(green) growing microcolonies from a strain of the yeast deletion collection containing the knockout of YFR054C, an open reading frame with dubiousfunction. Colony area as determined by automated image processing (solid lines, diamonds) and cell number as determined by a manual count (dashedlines, triangles) are plotted over time. Colonies generally display log-linear growth over the duration of the experiment. (D) Isogenic cells display a widedistribution of growth rates. A histogram of growth rates (grey) and a cumulative growth rate distribution (orange) of a population of ,17,000 isogenicmicrocolonies of the YFR054C knockout. (E) Gene deletions alter both growth rate mean and variance. Cumulative growth rate distributions of strains fromthe yeast deletion collection: two knockouts of open reading frames with dubious function (YFR054C and YHR095W), two knockouts that are unable to growwithout mitochondrial function (petite-negatives PET9 and YME1), and nine knockouts with diverse functions that have been shown to result in cell-to-cellheterogeneity in protein expression or morphology (SWA2, DIA2, KEM1, SNF6, RAD50, HTZ1, SCP160, BEM1, and NOT5). Note that a steep slope for thecumulative growth rate distribution indicates a low variance in growth rate.doi:10.1371/journal.pbio.1001325.g001
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 4 May 2012 | Volume 10 | Issue 5 | e1001325
growth as a petite [50,51] and compared these to growth
distributions of control strains of the same genetic background
but with a deletion of a dubious open reading frame (Figures 1E
and S5). Petite-negative strains generally contain as many or more
slow-dividing microcolonies than controls, suggesting petites are
not a major component of slow-dividing microcolonies in our
assay.
Lastly, we considered a high mutation rate as a possible source
of slow growth. Based on mutation accumulation experiments, the
spontaneous mutation rate in S. cerevisiae is estimated to be
,0.003–0.006 per cell per DNA replication when mutations to
homopolymeric runs are excluded [52,53], and ,0.3 when
mutations to homopolymeric runs are included [52]. Assuming
that each mutation results in an observable difference in growth
rate in our assay, mutation rates of this magnitude could explain a
large fraction of the growth rate variation. However, the
deleterious mutation rate is expected to be far lower than the
spontaneous mutation rate, and orders of magnitude below the
number of slow-growing colonies we observe. Indeed, in S. cerevisiae
the rate of fitness-altering mutations has been estimated to be
1.3761024 per haploid genome per generation [54]. Moreover,
we show below that slow growth is reversible for both single cells
and cell populations, suggesting that a large component of growth
rate heterogeneity is metastable and epigenetic in nature.
Mutations Alter the Variance in Growth RateWe have shown that wild-type yeast populations grown in a
benign environment contain a wide distribution of growth rates, a
property likely to impact population fitness in both static and
fluctuating environments. A static environment favors low
variance in growth rate, as the long-term population growth rate
of a single genotype is its geometric mean, which weighs lower
values of a distribution more heavily [55]. In contrast, a fluctuating
environment can favor high variance, if growth rate correlates
with stress survival [17]. The variance of the growth-rate
distribution is therefore an important evolutionary parameter,
but one that is invisible to standard, population-level measure-
ments of growth rate. To examine whether mutations can alter the
variance in growth rate, we selected candidates from previous
studies that had shown that deletions in gene products involved in
chromosome organization or those with a large number of genetic
and physical interactions increase the cell-to-cell heterogeneity in
gene expression or morphology [3,29,30,56]. We find that the
variance in growth rate can also be modulated by these deletions
(Figure 1E; Table S1). For example, deletion of histone variant
HTZ1 or the protein scaffold BEM1 results in a greater than 4-fold
increase in slow-growing microcolonies (operationally defined as
microcolonies growing at less than half of the population median)
when compared to control deletions of dubious open reading
frames. Interestingly, some gene deletions decrease the growth rate
variance. For example, deletion of SWA2, a gene that encodes a
product involved in clathrin-dependent vesicular transport, and
NOT5, a gene that encodes a global transcriptional regulator, each
result in a greater than 2-fold decrease in the number of slow-
growing microcolonies. There does not appear to be a trivial
relationship between the mean growth rate and variance (Figure
S6). A deletion resulting in a reduced mean growth rate can result
in an increased (DIA2, RAD50, HTZ1, BEM1) or decreased
(SWA2, SNF6) variance when compared to a deletion of a dubious
open reading frame.
Tsl1 Is a Marker for Slow Growth in a Benign EnvironmentTo allow for further investigation into the nature of growth
heterogeneity, we next sought to identify a molecular marker of
slow-dividing cell subpopulations. We reasoned that such a marker
would have at least two general characteristics: (1) a correlation
between its expression level and growth rate, and (2) high cell-to-
cell variation in its expression to match the observed variation in
growth rate. Genome-wide expression profiling of cells grown at
different growth rates in nutrient-limited chemostats has revealed a
large number of genes whose transcript levels correlate with
growth rate, no matter what the limiting nutrient [18]. However, a
correlation between a gene’s average expression level and the bulk
growth rate might merely indicate that the gene is part of a
generalized stress response. Indeed, genes whose transcript levels
correlate with growth rate overlap significantly with those that are
induced as part of a general environmental stress response [18].
To identify candidates among these genes that might be relevant
to growth heterogeneity under constant, benign conditions, we
therefore cross-referenced the growth-correlation data with data
on cell-to-cell variation in each protein’s abundance, as measured
by flow cytometry of cells engineered to encode a GFP fusion
protein at the corresponding endogenous gene [4]. Plotting these
two measures revealed several gene products that anti-correlate
with the population growth rate and that also exhibit a large
amount of cell-to-cell variation in protein levels under benign
conditions (Figure 2A). Using strict cut-offs for the growth-
correlation and protein-variation datasets, we identified 78
candidate markers of cell-to-cell variation in growth rate (Materials
and Methods) (Table S2).
We next investigated the candidate markers of cell-to-cell
variation in growth rate for enrichment in gene ontology (http://
www.geneontology.org) process, function, and component terms
(Table S3). As a group, the candidates appear to be involved in
energy storage or mobilization. Specifically, candidates are highly
enriched for mitochondrial genes in the proton-transporting ATP
synthase complex (p,961025) and genes involved in the
metabolism of the disaccharide trehalose (p,0.002). Trehalose is
synthesized by a trimeric complex consisting of two enzymatic
subunits, Tps1 and Tps2, and one of two interchangeable
cofactors, Tps3 and Tsl1 [57,58]. Among these, Tps1, Tps2,
and Tsl1 were identified as candidates in our screen, with Tsl1
ranking especially high for both the growth correlation and protein
noise datasets (Figure 2A). As a class, genes involved in trehalose
biosynthesis are highly over-represented among those whose
expression levels negatively correlate with growth rate and that
are induced by heat shock [20]. Both trehalose and Tsl1 appear to
be correlated with a stress-resistant cell state in yeast. Expression
levels of Tsl1 and bulk trehalose content remain relatively low
during exponential growth but rise rapidly as cells reach saturation
and become more stress resistant [58,59]. Trehalose is thought to
preserve protein folding under stress [60], and indeed cellular
trehalose content correlates with resistance to various forms of
stress, including heat, freezing, desiccation, and high ethanol
content [60–63]. Consistent with a direct role for Tsl1 in stress
resistance, deletion of TSL1 results in increased sensitivity to killing
by high ethanol concentrations [61]. Taken together, these data
suggest that Tsl1 might not only serve as a marker for a slow-
growing, stress-resistant cell state, but might also be an important
component of heterogeneity-dependent stress resistance. We
therefore chose to focus further examinations on the role of Tsl1
in bet hedging.
To determine if TSL1 expression correlates with individual slow
growth phenotypes in a non-stressful environment, we simulta-
neously monitored microcolony growth and green fluorescent
protein (GFP) fluorescence of cells encoding a Tsl1-GFP fusion
protein at the endogenous TSL1 locus [64]. As mentioned, Tsl1
abundance increases at saturation [58]. To avoid the possibility
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 5 May 2012 | Volume 10 | Issue 5 | e1001325
that variability in exit from stationary phase could confound our
results, we maintained cells in logarithmic growth for a minimum
of 24 h prior to any measurements. Consistent with previous flow-
cytometry data [4], the expression of Tsl1 varies between cells
(Figure 2B). An examination of individual microcolonies suggests
that, as predicted, Tsl1-GFP fluorescence correlates negatively
with cell-division rate. Figure 2B shows that cells undergoing few
or no cell divisions over the course of 8 h are highly Tsl1-GFP
fluorescent. Although GFP expression level and growth status tend
to persist within a cell lineage, they can change. Microcolonies
founded by a fast-dividing cell occasionally produce a highly
fluorescent cell with a low cell-division rate (Figure 2B). Cells can
switch in the opposite direction as well: a highly fluorescent cell
with a low cell-division rate can produce low-fluorescence fast-
growing progeny (Video S3). In general, slow-dividing cells appear
to be larger than fast-dividing cells (Figure 2B), suggesting that
these cells might have altered the influence of cell size on the Start
transition in late G1 [18,65].
The connection between high Tsl1-GFP fluorescence and low
cell-division rate, which we observe in individual cases such as that
shown in Figure 2B, holds as a general trend across many
microcolonies tracked in our assay. To control for alterations in
Tsl1 abundance that may be caused by differences in cell size, we
measured the Tsl1-GFP intensity per unit area of each microcol-
ony (Figures 2C and S7). A negative correlation between Tsl1
abundance and microcolony growth rate is observed across the
range of growth rates (Figures 2C and S8), indicating that Tsl1 is a
general marker of growth state rather than a marker for only
Figure 2. Tsl1 protein content marks slow growth. (A) A bioinformatic screen for candidates marking slow-growing cells. The correlation ofmRNA expression with the bulk population growth rate [18] is plotted against the protein-expression noise (the extent of cell-to-cell variation inexpression, DM in synthetic dextrose media from [4]) for each gene. 78 noisy genes that anti-correlate with the growth rate are in the upper leftquadrant (dashed lines). Among these are three subunits of the trimeric trehalose synthase complex (green circles). (B) Time-lapse microscopy of cellsexpressing Tsl1-GFP under the endogenous TSL1 promoter. Three time points of Tsl1-GFP fluorescence overlaid onto bright-field (left) and 1.67-foldmagnified views of bright-field (top right for each time point) and GFP fluorescence (bottom right for each time point) of three colonies from thefield. Arrows indicate the emergence of a morphologically distinct, slowly dividing, Tsl1-GFP fluorescent cell within a colony. (C) Tsl1 abundancecorrelates with growth rate. Top: A histogram of the specific growth rates of TSL1-GFP cells. Colors indicate bins used in bottom. Bottom: Tsl1-GFPfluorescence intensity per unit area of colonies binned by growth rate. Error bars indicate standard error of the mean (SEM); p-values are acomparison to all colonies; Wilcoxon-Mann-Whitney test: *, p,0.01; ***, p,1610210.doi:10.1371/journal.pbio.1001325.g002
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 6 May 2012 | Volume 10 | Issue 5 | e1001325
extreme slow growth. Also of note is that the correlation is specific
to Tsl1, not a generic property of any protein with variable
expression. Expression of the control protein Tma108-GFP, which
has a similar average abundance as Tsl1 [66] and is highly variable
from cell to cell [4], shows no correlation with microcolony growth
rate (Figure S9).
Slow Growth and Tsl1 Abundance Predict Resistance toHeat Killing
Having shown a correlation between Tsl1 abundance and
growth rate at the individual microcolony level, we next assayed
for differential susceptibility of microcolonies to heat killing. Tsl1-
GFP cells were grown normally in our microcolony growth assay
for 6 h (producing microcolonies of 1–20 cells), heat shocked
under conditions that kill most cells, and placed back under the
microscope for an additional 14–20 h of observation (Videos S4
and S5). Figure 3A shows a typical result: a highly fluorescent cell
in a slow-growing microcolony survives heat shock, undergoes one
or two cell divisions at a slow rate, and then produces fast-growing
progeny. Again, this individual case is representative of a general
relationship. Microcolonies with a higher Tsl1 content are
significantly more likely to contain a survivor, as shown by a plot
of the survival frequency of microcolonies binned by the Tsl1-GFP
fluorescence prior to heat shock (Materials and Methods)
(Figure 3B).
We next asked if Tsl1 is directly involved in heterogeneity-
dependent heat resistance or instead acts only as a marker of
resistant cells. We generated a genotypically similar TSL1
knockout strain, by replacing the coding sequence of the Tsl1-
GFP fusion protein with that of the fluorophore mCherry, to
compare the heat killing susceptibility of the TSL1D-mCherry
strain to the TSL1-GFP strain (Figure 3C). Multiple logistic
regression was used to isolate the effects of growth rate and TSL1
genotype on survival (Materials and Methods). Independent of
genotype, growth rate before heat shock is a major determinant of
survival, with slower growing microcolonies being more likely to
Figure 3. Growth heterogeneity is a stress survival mechanism. (A) Slowly growing, Tsl1 abundant cells survive heat shock. Time-lapseimages of Tsl1-GFP fluorescence overlaid onto bright-field (left) and Tsl1-GFP fluorescence of the right-most colony (right) before and after heatshock. Colonies are grown for 5 h to monitor growth rate and Tsl1-GFP fluorescence, heat shocked for 70 min at 60uC, resulting in massive cell death,and monitored for growth for 13 h following heat shock to identify colonies that contain at least one surviving cell. Bright-field and fluorescentimages of the entire field at each time point are shown in Video S4. (B) Tsl1-GFP fluorescent colonies are more likely to survive heat shock. Percentageof colonies that contain at least one cell that survives heat shock binned by Tsl1-GFP fluorescence. p-Values are a comparison to all colonies, Fisher’sexact test: **, p,161025, ***, p,1610210. (C) TSL1 contributes to survival in slow-growing colonies. Percentage of colonies that contain at least onecell that survives heat shock binned by growth rate for chimeric TSL1-GFP (green) or TSL1 replaced with mCherry (grey) at the endogenous TSL1 locus.Both growth rate and TSL1 genotype significantly affect survivorship (multiple logistic regression, p,10228 and p,0.01, respectively). (D) TSL1contributes to population resistance to acute heat shock. Survival of a strain containing a gene deletion of TSL1 (green) or a control dubious openreading frame (YFR054C, grey) as measured by plating on agar following heat shock of cell suspensions. Student’s t test of arcsin transformed data;**, p,161024. Error bars indicate SEM.doi:10.1371/journal.pbio.1001325.g003
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 7 May 2012 | Volume 10 | Issue 5 | e1001325
contain a survivor (multiple logistic regression, p,10228). Because,
prior to the heat shock, slow-growing microcolonies produce far
fewer cells than do fast-growing microcolonies, the difference in
survival per cell is necessarily greater than the differences reported
in our microcolony survival assay. In support of a direct role of
Tsl1 in heterogeneity-dependent stress resistance, functional Tsl1
improves survival when controlling for growth rate (multiple
logistic regression, p,0.01) (Figure 3C). The median growth rate
of TSL1D-mCherry populations is slightly reduced compared with
TSL1-GFP populations (Figure S10) and thus TSL1D-mCherry
populations would be expected to have more survivors if
survivorship is independent of TSL1 content. However, TSL1-
containing cells are slightly more likely to survive heat killing even
without controlling for the effects of growth rate on survival
(Figure 3C). One possibility to explain differential survival between
the TSL1-GFP strain and the TSL1 deletion is that TSL1 is an
important component of an induced heat shock response rather
than a component of a bet-hedging mechanism that renders a
proportion of cells heat resistant prior to any environmental shift.
To test this possibility, we compared the survival upon extremely
Figure 4. Cell sorting by Tsl1-GFP content alters growth rate distributions and heat shock survival. (A) The distribution of Tsl1abundance does not appear bimodal. Histogram of single-cell Tsl1-GFP intensity as measured by fluorescence-activated cell sorting. Inset: the right-hand tail of the main figure. Cells with the top 0.1% (dark green) and next 0.1 to 1% (light green) Tsl1-GFP fluorescence were sorted for downstreamanalysis. (B) Survival of sorted populations following heat shock of a liquid suspension as measured by plating on agar plates. p-Values are acomparison to the unsorted population (purple), Student’s t test of arcsin transformed data: *, p,0.01; **, p,161025. Error bars indicate SEM. (C)Sorting for Tsl1-GFP abundance transiently alters growth rate distributions. Left: cumulative growth rate distributions of sorted cells (dark and lightgreen), cells passed through the cell sorter but unsorted (purple), and cells not passed through the cell sorter (orange). Right: Samples of the samesorted or unsorted cell populations after ,42 generations of growth.doi:10.1371/journal.pbio.1001325.g004
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 8 May 2012 | Volume 10 | Issue 5 | e1001325
acute stress between a TSL1 knockout from the yeast deletion
collection and a second strain from that collection with a deletion
of a dubious open reading frame (YFR054C). Performing a 2-min
heat shock at 60uC in a small volume of liquid medium followed
by plating on agar to count survivors (Materials and Methods), we
find that TSL1 directly contributes to heat resistance under these
conditions in which an induced response is unlikely to be relevant
(Figure 3D).
Continuous Distributions Underlie ProbabilisticSusceptibility to Heat Killing
Having established TSL1 as both a predictor of susceptibility to
heat killing and an important component of the survival
machinery, we next sought to characterize the distribution of
TSL1 expression in yeast populations and how this distribution
relates to survival. As discussed previously, in bacteria, bistable
gene expression patterns underlie several binary phenotypic states
thought to act as bet-hedging mechanisms [8–14,24,67,68]. Thus,
levels of certain proteins show a bimodal distribution across cells.
Using flow cytometry to measure cellular Tsl1-GFP fluorescence,
we observe a continuous distribution in Tsl1 abundance rather
than the bimodal distributions characteristic of bistable bacterial
systems (Figure 4A). Sorting cells into discrete bins at the high end
of the Tsl1-GFP fluorescence distribution, then subjecting these
groups of cells to heat shock reveals that Tsl1 abundance predicts
survival in a graded or probabilistic manner rather than a binary
manner: the higher the level of Tsl1-GFP, the higher the chance of
survival (Figure 4B).
Taken together, observations of continuous or graded distribu-
tions in Tsl1 abundance (Figure 4A), growth rate (Figure 2C), and
stress survival (Figure 4B) suggest that populations of yeast might
contain a continuum of metastable epigenetic cell states that each
confer a different fitness in a given environment. This hypothesis is
supported by growth-rate distributions derived from cells sorted by
Tsl1 abundance. Cells sorted for higher Tsl1-GFP content yield
growth-rate distributions containing more slow-growing microcol-
onies (Figure 4C). If the altered growth-rate distribution of the cells
sorted for high Tsl1-GFP fluorescence reflects selection of a subset
of metastable cell states, then prolonged culturing of a population
founded by the sorted cells should result in a distribution similar to
the initial unsorted population, which is presumably a steady-state
distribution. The altered growth-rate distribution is indeed
transient. After 42 generations of growth, a population founded
by sorted cells has a growth-rate distribution that is indistinguish-
able from that of one founded by unsorted cells (Figure 4C, right).
Replicative Age Correlates with a Tsl1-Abundant CellState
As discussed previously, a combination of stochastic and
deterministic influences is likely to underlie the continuous and
graded distributions we observe here. Several characteristics of
stress-resistant cells led us to hypothesize that replicative age (the
number of cell divisions an individual cell has undergone) could be
a deterministic factor underlying yeast bet hedging. For example,
both old cells and stress-resistant cells have an increased cell size,
altered cellular morphologies, and a slowed cell cycle (Figure 2B)
[69–71]. To test this hypothesis, we stained TSL1-GFP cells with
wheat-germ agglutinin (WGA)-tetramethyl rhodamine isothiocya-
nate (TRITC), a fluorescent marker that specifically stains bud
scars, and measured single cell correlations in GFP and TRITC
fluorescence by flow cytometry. Older cells show higher levels of
TRITC fluorescence because each cell division leaves an
additional bud scar [72]. As predicted, cells that abundantly
express Tsl1 also show high levels of WGA-TRITC fluorescence
(Figure 5A). An alternative explanation is that cell states with high
Tsl1 abundance more efficiently take up the WGA-TRITC stain,
and the observed correlation is due to differences in staining rather
than replicative age. To test this possibility, we sorted cells for high
Figure 5. Old cells are Tsl1-GFP abundant. (A) TSL1-GFP yeast stained with the bud scar stain WGA-TRITC are passed though a cell sorter tomonitor co-fluorescence. Shown is the WGA-TRITC fluorescence of cells binned by Tsl1-GFP fluorescence. p-Values are a comparison to all cells,Wilcoxon-Mann-Whitney test: ***, p,1610210. (B) Sorted WGA-TRITC stained TSL1-GFP cells are counted for bud scars. Shown is the cumulativepercentage of all cells (grey) and cells in the top 1% Tsl1-GFP fluorescence bin (green). The 1% Tsl1-GFP cells have significantly more bud scars,Wilcoxon-Mann-Whitney test, p,161027. (C) Population demography accounts for some expression ‘‘noise.’’ Shown is the protein-expression noise(DM in yeast permissive dextrose media from [4]) of genes binned by logarithm of their age expression ratio, the average expression in young cellsover the average expression in old cells [73]. Error bars indicate SEM; p-values are a comparison to all colonies; Wilcoxon-Mann-Whitney test:*, p,0.05; **, p,0.001.doi:10.1371/journal.pbio.1001325.g005
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 9 May 2012 | Volume 10 | Issue 5 | e1001325
Tsl1-GFP content and compared the number of bud scars in
this subpopulation to an unsorted population by performing
manual bud scar counts. In further support of an age dependence
of Tsl1 expression, we find that cells with abundant Tsl1 tend to
have more bud scars (p,1027, Wilcoxon-Mann-Whitney test)
(Figure 5B).
Replicative Age Contributes Significantly to Protein-Expression Variation
The finding that variation in replicative age partially underlies
heterogeneity in TSL1 expression (and presumably heterogeneity
in growth rate and stress resistance) led us to hypothesize that
population demography might underlie a significant fraction of
protein-expression variation generally thought to be a conse-
quence of extrinsic noise. To test this hypothesis, we used data
from an existing microarray study that measured differences in
expression between young (one to three generations) and old (16–
18 generations) cells [73]. We then compared these expression
differences to data on cell-to-cell variation in each protein’s
abundance [4]. These abundances were measured by flow
cytometry of cells engineered to encode a GFP fusion protein at
the corresponding endogenous gene and therefore capture both
intrinsic and extrinsic noise, although a major source of extrinsic
noise (the cell cycle) was minimized by gating the cells by size and
complexity of shape [4]. Plotting the cell-to-cell variation in
expression of genes binned by their age expression ratio (AER, the
mean expression in young cells divided by the mean expression in
old cells) reveals that cell age does indeed contribute significantly
to protein-expression variation (Figure 5C, Wilcoxon-Mann-
Whitney test). Transcripts that become over- or under-expressed
in old cells tend to result in protein levels that are more variable
across cells in exponential growth. The absolute log AER explains
approximately 1% of the variation in protein-expression variation,
which is on par with other significant contributors to protein-
expression variation, including mRNA half-life, ribosomal density
of mRNA, and translation rate per mRNA [4].
Discussion
We show that: (1) clonal yeast populations contain a wide and
continuous distribution of growth rates when cultured in a benign
environment; (2) growth differences are transient and reversible
over the course of a few generations; (3) mutations can alter the
mean and variance of the steady-state growth rate distribution; (4)
Tsl1 is a marker for the slow-growing cells within an exponentially
growing population; (5) Tsl1 abundance and slow growth predict
resistance to heat killing; (6) TSL1 is an important component of
heterogeneity-dependent heat shock survival; (7) Tsl1-abundant
cells tend to be of higher replicative age; and (8) replicative age is
likely to underlie a fraction of gene expression heterogeneity for
many gene products besides Tsl1.
These results describe a bet-hedging phenomenon in yeast that
might be an adaptation to life in an unpredictably varying
environment (Box 1). As is true in descriptions of bacterial bet
hedging and persistence [16], slow growth is a crucial predictor of
stress survival in yeast. Both bacteria and yeast appear to be
maximizing population fitness by balancing fast growth in good
conditions with bet hedging against bad ones [17].
Yet, some crucial differences between bacterial and yeast bet
hedging appear prominent. One difference appears to be in the
nature of the heterogeneity underlying bet hedging. Single-cell
observations in bacteria suggest that persisters and non-persisters
constitute binary growth states that predict survival in an all-or-
none fashion [16]. That is, bacterial persisters generally survive
and non-persisters generally perish in stress. We find that yeast
populations contain a continuous rather than bimodal distribution
of growth states and that these states predict survival in a
probabilistic manner. That is, the slower a yeast cell grows, the
greater its probability of surviving stress. Although the mechanism
of bacterial persistence has yet to be elucidated, persisters and non-
persisters are thought to interconvert through a stochastic
mechanism [16], as is true for the vast majority of characterized
bacterial two-state systems [8–14]. In yeast, differences in growth
and survival appear to be due to a more complex combination of
stochastic and deterministic factors. Taken together, these results
suggest that bet hedging in yeast is a consequence of a spectrum of
metastable inheritable epigenetic states that confer differential
fitnesses across environments.
The processes underlying interconversion between epigenetic
states, and the different phenotypes associated with these states,
are of great importance not just for yeast but also in metazoan
development and disease. Interconverting epigenetic states have
been shown to underlie phenomena as diverse as antibiotic
resistance [16], stem cell reprogramming [74], and cancer
progression [75–77]. For example, recent work has shown that
rare cells within a melanoma tumor divide slowly but give rise to
highly proliferative daughter cells, and vice versa [78]. This
behavior can be thought of as a bet-hedging mechanism, and likely
contributes to the poor long-term performance of chemotherapies
that target fast-dividing melanoma cells [78].
Current theoretical models of bet hedging focus on the
dynamics of two-state systems [17,16]. Our results and recent
work in human cancer cell lines [77] suggest that future models
must account for a distribution of multiple cell states and the
transitions between them [79,80]. Interconversion between
multiple Tsl1-abundance and growth states presents an experi-
mentally tractable system that can be exploited to test and
parameterize such models. For example, sorting cells by Tsl1
abundance and following changes in growth rate distributions over
time might allow for theoretical estimates of the number cell states
and the transition rates between them [77]. Additionally, because
the microcolony growth and survival assay presented here relies on
simple microscopy and image analysis routines, these methods
could be relatively easily exported to the above-mentioned cell
culture systems to provide additional quantitative measures of
metazoan multi-stability and bet hedging.
A correlation between growth and the deterministic factor of
replicative age has been previously noted, with increasing age
resulting in slower progression through the cell cycle until no more
cell cycles can be completed [71]. Here we show that replicative
age also correlates with Tsl1-abundant, presumably stress-resistant
cell states. We note, however, that replicative age does not appear
to be the sole determinant of slow growth, Ts1-abundance, or
stress resistance. Both slowed growth and high fluorescence of
TSL1-GFP cells persist in newborn cells and their daughters
(Figure 1B, 1C; Video S3). Additionally, we observe newly born
cells surviving heat shock (Videos S4 and S5). A more likely
possibility is that both the deterministic factor of replicative age
and stochastic mechanisms contribute to stress resistance, although
more research is required to establish causal links.
The possibility that old age contributes to stress resistance
provides a particularly compelling bet-hedging mechanism: an old
cell with few remaining cell cycles maximizes its contribution to a
clonal population if a cell cycle is completed after a stressful event
that results in a mass killing of the younger, fast-growing cells.
Thus, a slowed cell cycle in old cells—and, with high probability,
their few remaining progeny, as implied by inheritance of TSL1
abundance and slow growth from mother to daughter (Video
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 10 May 2012 | Volume 10 | Issue 5 | e1001325
S3)—might be selected to maximize bet hedging. In this scenario,
the influence of age, although independent of the environment,
could nonetheless be probabilistic, if the age signal or its transduction
is noisy. An alternative possibility is that older cells have accumulated
minor stresses throughout their lifetimes, so that induction of TSL1
and other genes represents a genuine stress response despite the
benign environment. If true, this possibility would still represent a
bet-hedging mechanism, because the induced response protects
against a subsequent, unpredictable, and acute stress.
We have previously shown that a large number of gene deletions
result in a decreased phenotypic robustness, increasing the cell-to-
cell heterogeneity in morphology [30]. Here we show that some of
these mutations also alter the growth rate distribution in a benign
environment, often resulting in a greater variance in growth rates
with proportionally more slow-dividing cells. It is unclear whether
the large number of slow-dividing cells in the distributions of these
single-deletion strains represent, as they do in wild-type popula-
tions, meaningful bets that are more fit in harsh environments, or
instead represent unfit cell states in any environment. Yet, the
possibility that a large number of mutations could result in
increased fitness in harsh environments presents a dynamic picture
of the tension between bet hedging and robustness in yeast. That
is, selection for a robust phenotype in a given environment (i.e.,
the fittest phenotypic state) is countered by selection for distributed
phenotypes (i.e., multiple phenotypic states that constitute a series
of bets on changing environments) [81]. When the environment is
not constant and when slow growth in a benign environment
confers resistance to an acute stress, then the growth rate
distribution of a mutant will be more informative about fitness
than the mean growth rate. The high-throughput microcolony
assay of growth and stress survival offers a way to explore these
distributions systematically using yeast gene-deletion strains or
strains segregating natural variation. Another way to explore the
tension between robustness and bet hedging would be to test the
expectation that organisms evolved in fluctuating environments
should exhibit a wider distribution of growth rates than those
evolved in static environments.
We have shown here that Tsl1, a protein involved in the
synthesis of trehalose, is both a marker and an important
component of a stress-resistant cell state. Trehalose appears to
function as a general stress protectant across biological kingdoms,
approaching 20% of the dry weight of stress-protected organisms
such as yeast and nematodes, which regularly encounter harsh
conditions [59]. Thus, it is quite plausible that the bet-hedging
mechanism described here will provide mechanistic insights into
ecological adaptation in a wide range of organisms, as well as into
how pathogenic eukaryotes, such as C. albicans [82] or indeed
strains of S. cerevisiae [83], colonize humans and evade therapeutic
agents. Identification of Tsl1 as marker for stress-resistant cell
states in yeast will be of great value to elucidating the molecular
mechanisms underlying persistence, an endeavour that has been
elusive in bacterial models [21]. For example, comparisons of the
gene expression profiles between cells sorted for abundant Tsl1-
GFP and unsorted populations will provide a list of candidate gene
products involved in heterogeneity-dependent stress resistance.
These candidates can subsequently be tested for correlation to or
necessity within a stress-resistant cell state using methods similar to
those described here for Tsl1.
Materials and Methods
Yeast Strains and CloningHaploid deletion strains were converted from the diploid
BY4743 YKO magic marker strains (Open Biosystems, MATa/
a ura3D0/ ura3D0 leu2D0/leu2D0 his3D1/ his3D1 lys2D0/LYS+met15D0/MET15+ can1D::LEU2+-MFA1pr-HIS3/CAN1+ xxx::
kanMX/XXX+) as described [84]. Tsl1-GFP yeast (MATa
his3D1 leu2D0 met15D0 ura3D0) are part of the yeast-GFP
collection [64] and were purchased from Invitrogen. The TSL1D-
mCherry strain was constructed directly from the TSL1-GFP
strain by replacing the coding region of the genomic TSL1-GFP
chimera with that of mCherry and the selectable marker, NatMX,
through homologous recombination [85]. The mCherry-NatMX
insert was amplified by PCR from the pCZ-Nat plasmid (GenBank
accession number JN580989) using the primers (59R39)
CAAACAAAGCAAAGAATACAATAGCAACGCAAGATCA-
ACACAATGGTGAGCAAGGGCGAGGAGGA and AAGTT-
CATACCCAAGAAAATTAAAATTTTAAAATGGTAAAATT-
TATGAGCTCCAGCTTTTGTTCC. The PCR product was
extended for homologous recombination using the primers
CCGTGTCATTGCACATCCACCCACCCGTCGATTAAA-
AAACCAAACAAAGCAAAGAATACAATAGCAACG and TA-
GAATTGATATATAATAAGCAGTTGAAAATAAAAGTTC-
ATACCCAAGAAAATTAAAATTTTAAAATGG. Transforma-
tion of the PCR construct was performed with lithium acetate, as
described [86], and homologous recombinants were selected for
incorporation of NatMX with nourseothricin. Proper integration
was confirmed by sequencing.
Cell PreparationFor all strains, a single colony was selected and grown overnight
in YPD to generate a frozen stock. Frozen stocks were struck onto
YPD plates at a high density and populations from the streak were
used to initiate experiments. Growth rate and survival assays were
preformed in synthetic complete liquid medium or on synthetic
complete plates. Deletion strains were allowed to reach saturation
in liquid culture. A day prior to plating, saturated cell cultures
were diluted 1:60 and grown overnight to saturation. On the day
of plating, cultures were again diluted 1:60 and allowed to reach
early logarithmic phase by growing for 3–4 h with shaking at
30uC. Because TSL1 expression increases for all cells during late
log phase and saturation [58], the TSL1-GFP and TSL1D-
mCherry strains were instead maintained in early- to mid-log
phase for at least 24 h prior to any experiments. We estimate ,50
generations of growth between a single cell bottleneck and growth
rate measurements for single gene deletions strains and ,60
generations for TSL1-GFP and TSL1D-mCherry strains.
MicroscopyCells were sonicated for 90 s on high in a Diagenode Bioruptor
water bath, counted using a hemocytometer, and diluted to a final
concentration of 5–206103 cells/ml. Glass-bottomed 96-well
plates (Matrical MGB096-1-2-LG) were coated with 200 ml of
200 mg/ml Concanavalin A (Type V, Sigma) for 2–6 h. Wells
were washed once with 200 ml of water and 400 ml of cells were
plated per well. Plates were sealed with an optically clear film
(Axygen PCR-SP) and spun at 360 g for 2 min. Before placing the
samples on the microscope, the bottom surface of the 96-well plate
was dusted using compressed air to remove any particles that may
interfere with the Nikon Perfect Focus System (PFS). Micrographs
were captured on a Nikon TE2000e microscope equipped with
PFS for infrared high-speed focusing and a fully automated stage
equipped with a full-stage environmental chamber. All images
were collected with a Nikon Plan Apo 106 (0.45 numerical
aperture) air objective using Nikon NIS Elements software to drive
stage movement and acquisition. Because NIS Elements readily
accepts externally written XML files for position and PFS control,
we created homemade R- and C-based scripts to assign plate
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 11 May 2012 | Volume 10 | Issue 5 | e1001325
position and focus coordinates that minimize stage travel time and
optimize PFS offsets over the plate surface. The environmental
chamber was set to 30uC at least 2 h before observation to prevent
heat gradients. Prior to image acquisition, a 45-min focusing
routine was performed to determine the optimal PFS offset for
each well. This focusing routine was necessary because the PFS
maintains focus on the plane between the bottom of the cover glass
and the air. Thus, alterations in thickness of the cover glass surface
result in images that are slightly out of focus, which we found can
have mild effects on measured growth rates (unpublished data).
Microcolonies begin growing during the focusing routine, thus
some microcolonies might contain two cells by the time an initial
image is taken. Microcolony growth was monitored by capturing a
micrograph of each field every hour. For routines that require only
bright field images, we were able to monitor ,3,000 fields in
parallel (,100,000 microcolonies at 26104 cells/ml) because each
image requires ,1 s to capture including stage travel time. Tsl1-
GFP fluorescence was captured for 4 s at 26 gain. The long
exposure time required for fluorescence measurements consider-
ably slowed our assay, allowing us to monitor ,360 fields (,3,000
microcolonies at 0.56104 cells/ml).
Automated Image AnalysisImage processing and microscope control routines were written
in Matlab, R, C, and shell scripts.
Focusing routine. For each well, bright-field images of five
fields at four different PFS offsets were captured. The five fields
map to the center and four corners of a rectangle that covers the
central 50% of the area of each well. The four PFS offsets are
spaced 7.5 mm apart, which allows at least one image to be
captured in the proper focal plane to determine the correct PFS
offset (the glass surface typically varies by ,30 mm over a plate).
To find the ideal optical plane for each field in a computationally
efficient manner, we took advantage of an idiosyncratic optical
property of yeast: ,10 mm below the ideal focal plane, yeast
appear large and white resulting in a histogram of image pixel
intensities with a sharp peak at the highest measured pixel
intensity. To find the ideal optical plane for each field, we found
the PFS offset of the image with maximal number of highest
intensity pixels and added 10 mm. For each well, the optimal PFS
offset of each of the five fields was averaged and this averaged
offset was used for all images captured in the next phase of the
experiment. Capturing 1,920 images, (96 wells)6(5 fields)6(4 PFS
offsets), requires ,40 min of microscope time and performing the
simple image analysis and calculation routine required about
5 min of computational time.
Growth rate analysis. Bright-field images were analyzed
continuously during image capture on a different dedicated
computer. The area of each micrograph covered by yeast
microcolonies was identified by taking advantage of the fact that
yeast tend to be both the lightest (high pixel intensities in the cell
center and outside the cell perimeter) and dimmest (low pixel
intensities at the cell perimeter) objects in the field. Thresholds
were applied to both high and low pixel intensities to create a pair
of black (not-yeast) and white (yeast) images for each optical field.
Each black and white image was then subjected to several rounds
of optical dilation and erosion to generate continuous white
microcolony objects. Non-yeast objects that were erroneously
identified as yeast in either the high or low threshold were
removed by performing an AND operation across the high and
low pixel threshold images. Thus, only objects that contain both
high and low intensity pixels, a property specific to yeast rather
than cellular debris or other precipitates, are counted as a
microcolony. Once all images from a time-series are collected,
microcolonies are aligned through time by centroid proximity.
When a microcolony physically touches a neighboring microcol-
ony or the edge of the image field, it is no longer tracked. Large
decreases in microcolony area, generally indicating an image
analysis failure, cause the software to discontinue tracking that
microcolony. Microcolonies that appear de novo in a later time
point (generally a single cell that floated away from a nearby
microcolony) are grouped with the nearest microcolony if they lie
within a distance of (0.65)6(length of the longest line that can be
drawn through the microcolony), otherwise they are ignored. For
each microcolony, the total recorded time of growth is measured
and microcolonies with less than five recorded growth time points
are ignored. When log(microcolony area) is plotted over time for
9 h of growth, in excess of 99.9% of microcolonies that underwent
at least one cell division (operationally defined as a specific growth
rate above 0.1) displayed linear correlations in excess of 0.9,
suggesting that cells are not limited for nutrients over the time of
observation.
Fluorescence quantitation. Tsl1-GFP fluorescence was
quantified by averaging the intensities in the fluorescent channel
of all pixels within the microcolony area of each microcolony as
determined in the above section.
Post-heat shock alignment. Removal and replacement of
the plate for heat shock causes slight shifts in the locations of each
microcolony, often causing errors in centroid-based microcolony
alignment through time. Thus, we added a routine to realign
microcolonies by searching across many fields on the plate for
microcolonies with distant neighbors. The average centroid
movement of these isolated microcolonies was used to calculate
total plate movement and realign all remaining microcolonies.
Reliable realignment required cells to be plated at a lower density.
Thus, all heat shock experiments were performed at 56103 cells/
ml (,eight microcolonies per field). We found that this lower
density did not alter growth rate distributions (unpublished data).
Microcolony survival. Microcolonies were labeled as survi-
vors if they grew by 400 or more pixels after heat shock (180 mm2
or ,eight cells) in 16 h of growth. Microcolonies were labeled as
non-survivors if they grew by less than 300 pixels after heat shock
(135 mm2 or ,six cells) in 16 h of growth. Microcolonies that did
not meet these criteria were ignored. Alternative values of the
change in area cut-offs for calling survivors or non-survivors or of
the length of growth allowed before assessing survival had no effect
on the conclusions of this paper (unpublished data).
Measures for Systematic BiasMicrocolony proximity measurements. The centroid of
each microcolony was calculated by averaging the centroid
positions of all time points for which the microcolony was tracked.
Proximity of each microcolony to its nearest neighbor was
calculated by measuring the minimum distance between a
microcolony and all other microcolonies in the field. Microcolo-
nies within 55 mm of the field edge were ignored for all
measurements. A small fraction of microcolonies that fall within
35 mm (4–8 cell lengths) of their nearest neighbor do have reduced
growth rates (Figure S3). Although it is possible that this skew is
due, in part, to local nutrient depletion, a more likely explanation
is a technical bias of the experiment: fast-growing close-neighbor
microcolonies merge with neighbors before sufficient time points
can be recorded to estimate an accurate growth rate (a minimum
of five time points is necessary in these assays), whereas slow-
growing close-neighbor microcolonies do not.
Position on the plate. To measure systematic bias that may
be caused by the position of a well on a 96-well plate or the
position of a microcolony within a well, we performed several
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 12 May 2012 | Volume 10 | Issue 5 | e1001325
replicates of growth with the identical strain in each well, a haploid
segregant of a dubious open reading frame knockout (YFR054C)
from the BY4743 YKO collection. We found no obvious bias
based on well position or position within a well. However, to be
conservative, we randomized the well position for all growth rate
measurement of strains from the YKO collection.
Manual Microcolony Cell CountsCell counts were performed on a haploid segregant of a dubious
open reading frame knockout (YFR054C) from the BY4743 YKO
collection. Microcolonies on which to perform manual cell counts
were selected at random from the following growth rate bins:
below two standard deviations (2 SD) from the mean population
growth rate, between 2 SD and 1 SD below the mean, between 1
SD below the mean and 1 SD above the mean, between 1 SD and
2 SD above the mean, and above 2 SD above the mean. Counts
did not try to distinguish budding cells that have not undergone
cytokinesis from two separate cells (i.e., a cell with a small bud was
counted as two cells). Microcolonies under ,100 cells generally
grew as a monolayer on the glass surface and cell counts correlated
extremely well with automated colony area measurements
(Figures 1C, S1, and S2). We did notice that automated
measurements slightly overestimated the cell number of the
slowest growing microcolonies when they became large (Figure
S2A) and slightly underestimated the cell number of the fastest
growing microcolonies when they became large (Figure S2E). For
both slow and fast growing microcolonies, the automated
measurements provide conservative estimates for the deviation
from the mean growth rate (i.e., slow-growing colonies are
measured as growing faster than their true growth rate), and thus
the automated growth rates were not adjusted. For microcolonies
over ,100 cells, we did notice some piling of cells on top of each
other resulting in automated colony area measurements underes-
timating the total number of cells (unpublished data). We therefore
limited all of our quantitative assays to colony sizes below 100 cells.
Single-Gene Deletion Growth Rate DistributionsIn each well of a 96-well plate, approximately equal cell
numbers of a single-gene deletion strain and an easily distinguish-
able GFP fluorescent control strain, FBA1-GFP [64], were grown
together for 9 h. The mean growth rate of the FBA1-GFP
microcolonies was used to normalize deletion strain growth rates
across different experimental wells. All reported single-gene
deletion distributions are the combined microcolony growth rates
from at least three replicate wells of a 96-well plate. Relative
growth rates reported in Figures 1E and S3 and Table S1 were
calculated by setting the mean growth rate of the control dubious
open reading frame deletions (YHR095W and YFR054C) equal to
one. Alternatively, growth rate distributions of deletions resulting
in petite-negative and dubious open reading frame control strains
(Figure S5) are reported as raw specific growth rates without
normalization.
Screen for Slow-Growth MarkersUsing microarray analysis of cells grown in nutrient-limited
chemostats, the regression slope for the transcriptional response to
changes in growth rate has been determined for all transcripts
[18]. The cell-to-cell variation, or noise, in protein level has been
quantified for a large number of genes using flow cytometry of
endogenously expressed GFP fusions and is summarized in a
measure called DM [4]. To identify gene products that might
mark cell-to-cell variation in growth rate, we set the following
thresholds: a growth rate slope of less than 22 and noise (DM in
synthetic dextrose medium) of greater than 5.
Gene Ontology EnrichmentGene ontology enrichment was calculated using the GO Term
Finder in the Saccharomyces genome database website (http://
www.yeastgenome.org/cgi-bin/GO/goTermFinder.pl) on Sep-
tember 7, 2011 using the default settings. Genes with a reported
value for both growth rate slope and noise in synthetic dextrose
medium were used as the set of background genes for statistical
comparisons. Reported p-values are corrected for multiple
hypothesis testing.
TSL1-GFP Growth Rate DistributionsAs is true for the single deletion studies, we observed only
nominal differences between replicate wells or days (unpublished
data). Thus, growth rate distributions and fluorescence correlation
studies for the TSL1-GFP strain and associated controls were
generated by pooling microcolony growth rates across a minimum
of 12 replicate wells and two experimental days. Growth rate
distributions associated with microscopy-based survival assays
were performed by pooling microcolonies from a minimum of 80
replicate wells over a minimum of two experimental days.
Heat ShockHeat shock of film-covered glass-bottomed micro-well plates
was performed by removing plates from the microscope and
sandwiching them between two pre-heated standard aluminum
heat blocks in a hydrated oven for 70 min at 60uC. Heat shock of
TSL1 and control knockout strains was performed in liquid
suspension for 2 min at 60uC. Heat shock of sorted and unsorted
TSL1-GFP strains was performed in liquid suspension for 6 min at
52uC.
Growth-Rate Binned SurvivalA TSL1D-mCherry control strain was constructed as a direct
descendant of the TSL1-GFP strain (see ‘‘Yeast Strains and
Cloning’’). This genetic manipulation resulted in a mild but
detectable decrease in mean population specific growth rate
(0.366 h21 for TSL1D-mCherry, 0.377 h21 for TSL1-GFP,
p,10210, Wilcoxon-Mann-Whitney test). Because TSL1 expres-
sion is generally low, requiring long exposure times and thus high
fluorescence background, TSL1D-mCherry and TSL1-GFP were
grown in a checkerboard pattern on separate wells on the same
plate, rather than in the same well, to avoid genotype miscalls. We
observed no obvious biases in growth rate or survival frequency
over a plate’s surface for either genotype and observed similar
survival patterns over four similar heat-shock experiments.
Multiple Logistic RegressionHeat-shock survival is a binary dependent variable, so multiple
logistic regression was used to test the effects on it of growth rate
(prior to heat shock) and genotype (TSL1-GFP vs. TSL1D-
mCherry). A full linear model including main-effect terms for
growth rate and genotype as well as an interaction between the
two was compared to reduced models using AIC, as implemented
in the glm and anova functions of R. The full model was not
significantly better than a reduced model with the two main effects
but without the interaction. Removing either main effect from this
no-interaction model made the model significantly worse. There-
fore reported p-values come from the model with both main effects
but no interaction.
Survival in Liquid SuspensionHeat shock of liquid suspensions was performed in triplicate.
Survival frequency was determined by counting the number of
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 13 May 2012 | Volume 10 | Issue 5 | e1001325
cells in liquid before heat shock and comparing this to the number
of colonies that grew on an agar plate following heat shock. p-
Values were determined by performing a Student’s t test on the
arcsin of the square root of the proportion surviving.
Cell SortingCells were sonicated for 90 s in a Diagenode Bioruptor water
bath prior to sorting. Cell sorting was performed on a FACSaria
(BD) sorter. The pulse width was used to separate individual cells
from cell clumps. FITC gates used to isolate cells with high levels
of Tsl1-GFP are shown in Figure 4A. Because most cells contain
low levels of Tsl1 (,2,000 molecules per cell on average [66]), the
sorter was not sensitive enough to sort cells in the bottom 85% of
the distribution. Sorted cells were immediately resuspended in
synthetic complete medium following sorting. A fraction of cells
were immediately plated for growth rate distribution and heat
shock survival analysis. A second fraction was grown in early- to
mid-log phase for 48 h, allowed to reach saturation, grown again
in early- to mid-log phase for 24 h, and plated for growth rate
distribution analysis. Assuming an average specific growth rate of
0.4 h21 for each sorted fraction, 76 h of log growth represents
,42 generations.
Replicative Age AnalysisTSL1-GFP cells were kept in logarithmic growth for a minimum
of 24 h, sonicated for 90 s in a Diagenode Bioruptor water bath,
washed once in PBS, fixed for 90 min in 3.7% formaldehyde, and
washed twice in PBS. Bud scar staining was performed for 15 min
in 1 mg ml21 TRITC-labeled WGA. All sorting was done using a
tight pulse-width gate to remove cell clumps from the analysis. For
co-fluorescence measurements, ,86105 cells were measured for
WGA-TRITC and Tsl1-GFP fluorescence. Data shown are from
one experiment. Replicate experiments yielded similar results. For
bud scar counts, cells were sorted until 104 cells were recovered.
Cells were pelleted, resuspended in 5 ml Vectashield (Vector
Laboratories), and mounted on a glass slide. Bud scars were
counted manually using a Nikon TE2000e epifluorescent micro-
scope and a 1006plan apochromat objective with a narrow focal
plane. Three sorts for each category were performed and bud scars
from ,100 cells were counted per sort. Similar bud scar
distributions were observed in all three sorts. Data shown are
the pooled counts from all sorts.
Supporting Information
Figure S1 Automated colony area measurements correlate with
cell number. Manual cell counts are plotted against colony area
measurements determined by automated image processing for
cells binned by growth rate for the strain of the yeast deletion
collection containing a knockout of YFR054C, an open reading
frame with dubious function. (A) below two standard deviations (2
SD) from the mean population growth rate, (B) between 2 SD and
1 SD below the mean, (C) between 1 SD below the mean and 1
SD above the mean, (D) between 1 SD and 2 SD above the mean,
(E) above 2 SD above the mean, (F) all counts from (A–E) plotted
together. The purple line indicates the linear regression of the
points.
(PDF)
Figure S2 Bland-Altman plot of automated and manual cell
counts. Manual cell counts are plotted against the difference
between cell count estimations on the basis of automated colony
area measurements and manual cell counts. Cells are binned by
growth rate for the strain of the yeast deletion collection
containing a knockout of YFR054C, an open reading frame with
dubious function. (A) below 2 SD from the mean population
growth rate, (B) between 2 SD and 1 SD below the mean, (C)
between 1 SD below the mean and 1 SD above the mean, (D)
between 1 SD and 2 SD above the mean, (E) above 2 SD above
the mean, (F) all counts from (A–E) plotted together. The purple
line indicates the mean difference and orange lines indicate the
95% confidence interval.
(PDF)
Figure S3 Reproducibility of the microcolony growth rate assay.
Eighteen replicate growth rate distributions are shown for six yeast
strains. Traces of the same color are from replicate wells on the
same microplate and traces of different colors are from replicate
experimental days. In addition to the genotypes shown, each well
contained an easily distinguishable fluorescent strain from the GFP
fusion collection (FBA1-GFP, Invitrogen) [64] that was used to
normalize growth rates for global differences between wells or
experimental days (Materials and Methods). Thus, growth rates
are reported as normalized values, with the mean FBA1-GFP
growth rate within each well used as the normalizing denominator.
(PDF)
Figure S4 Effect of colony proximity on growth rate. Box plot of
growth rates of colonies binned by the colony’s proximity to its
nearest neighbor for cells plated at a density of 26104 cells/ml.
Colonies that fall within 35 mm (4–8 cell lengths) of their nearest
neighbor do have reduced growth rates. This reduction may be due
to local nutrient depletion or a technical problem with measuring
the growth rates of closely spaced colonies (Materials and Methods).
Regardless of the cause, we ignored all colonies with a nearest
neighbor of less than 35 mm away in all our measurements.
Whiskers are 1.5 times the interquartile range from the box.
(PDF)
Figure S5 Petite-negative strains contain slow-growing colonies.
Cumulative specific growth rate distributions of six haploid
knockout strains from the yeast deletion collection. Plotted are a
control dubious open reading frame knockout (YFR054C, black)
and five knockouts unable to grow when mitochondrial function is
lost: YME1 (red), PET9 (green), MGR1 (blue), MGR2 (cyan), and
PDE2 (magenta).
(PDF)
Figure S6 Scatter plot of the means and standard deviations of
the microcolony growth rates of 13 knockout strains from the yeast
deletion collection. A linear regression (blue line) does not fit the
data well.
(PDF)
Figure S7 Tsl1-GFP fluorescence intensity per unit area of
colonies binned by growth rate for each time point during the first
6 h of growth. p-Values are a comparison to all colonies,
Wilcoxon-Mann-Whitney test: *, p,0.01; **, p,161025;
***, p,1610210.
(PDF)
Figure S8 Scatter plot of Tsl1-GFP fluorescence intensity per
unit area and specific growth rates of microcolonies of cells
expressing Tsl1-GFP under the endogenous TSL1 promoter. A
linear regression of the data is shown as a green line.
(PDF)
Figure S9 Fluorescence intensity per unit area of colonies
binned by growth rate for TSL1-GFP (left) or TMA108-GFP cells
(right). Error bars indicate SEM; p-values are a comparison to all
colonies; Wilcoxon-Mann-Whitney test: **, p,161025;
***, p,1610210.
(PDF)
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 14 May 2012 | Volume 10 | Issue 5 | e1001325
Figure S10 Cumulative specific growth rate distributions of
TSL1-GFP and TSL1D-mCherry microcolonies that contained at
least one cell that survived heat killing or no surviving cells.
(PDF)
Table S1 Characteristics of growth rate distributions of single
gene deletion strains.
(XLS)
Table S2 Candidate genes involved in bet hedging.
(XLS)
Table S3 Gene ontology enrichment of noisy gene products
whose average expression correlates with the bulk growth rate.
(XLS)
Video S1 A typical field of microcolony growth. Cells in
logarithmic growth are plated at a low density on a glass-bottomed
multi-well plate and low-magnification time-lapse bright-field
images are captured each hour (left). Custom-written software
tracks colony area over time (right). Colonies that touch each other
or the edge of the field discontinue being tracked. Notice an
extremely slow-growing dark-purple colony at the lower left of the
field.
(MOV)
Video S2 A video of the same colonies shown in Figure 1A.
(MOV)
Video S3 A highly fluorescent slow-growing cell produces low-
fluorescence fast-growing progeny. Bright-field (bottom) and
fluorescent (top) images of TSL1-GFP yeast grown for 9 h. Notice
that slow-growing cells on the right produce fast-growing progeny.
(MOV)
Video S4 Typical results of heat killing. Bright-field (left) and
fluorescent (right) images of TSL1-GFP yeast grown for 5 h, heat
shocked, and grown for another 13 h.
(MOV)
Video S5 A second field, as described in Video S4.
(MOV)
Acknowledgments
We thank David Gresham, Jonathan Guberman, Zemer Gitai, Bob
Johnston, Dmitri Petrov, Gavin Sherlock, Dan Kvitek, and Jared Wenger
for helpful discussions and advice on the manuscript. We thank Pui-Leng
Ip, Christina Lyman, Sandra Schnakenberg, Joshua Richardson, and
Yakov Pechersky for technical assistance.
Author Contributions
The author(s) have made the following declarations about their
contributions: Conceived and designed the experiments: MLS SFL.
Performed the experiments: SFL NZ. Analyzed the data: MLS SFL.
Contributed reagents/materials/analysis tools: MLS SFL. Wrote the
paper: MLS SFL.
References
1. Ko MS, Nakauchi H, Takahashi N (1990) The dose dependence ofglucocorticoid-inducible gene expression results from changes in the number
of transcriptionally active templates. EMBO J 9: 2835–2842.
2. Elowitz MB, Levine AJ, Siggia ED, Swain PS (2002) Stochastic gene expression
in a single cell. Science 297: 1183–1186.
3. Raser JM, O’Shea EK (2004) Control of stochasticity in eukaryotic geneexpression. Science 304: 1811–1814.
4. Newman JRS, Ghaemmaghami S, Ihmels J, Breslow DK, Noble M, et al. (2006)
Single-cell proteomic analysis of S. cerevisiae reveals the architecture ofbiological noise. Nature 441: 840–846.
5. Raj A, Peskin CS, Tranchina D, Vargas DY, Tyagi S (2006) Stochastic mRNA
synthesis in mammalian cells. PLoS Biol 4: e309. doi:10.1371/journal.-pbio.0040309.
6. Altschuler SJ, Wu LF (2010) Cellular heterogeneity: do differences make a
difference? Cell 141: 559–563.
7. Eldar A, Elowitz MB (2010) Functional roles for noise in genetic circuits. Nature
467: 167–173.
8. Maamar H, Dubnau D (2005) Bistability in the Bacillus subtilis K-state(competence) system requires a positive feedback loop. Mol Microbiol 56:
615–624.
9. Maamar H, Raj A, Dubnau D (2007) Noise in gene expression determines cellfate in Bacillus subtilis. Science 317: 526–529.
10. Smits WK, Eschevins CC, Susanna KA, Bron S, Kuipers OP, et al. (2005)
Stripping Bacillus: ComK auto-stimulation is responsible for the bistableresponse in competence development. Mol Microbiol 56: 604–614.
11. Veening J-W, Hamoen LW, Kuipers OP (2005) Phosphatases modulate the
bistable sporulation gene expression pattern in Bacillus subtilis. Mol Microbiol
56: 1481–1494.
12. Fujita M, Losick R (2005) Evidence that entry into sporulation in Bacillus subtilisis governed by a gradual increase in the level and activity of the master regulator
Spo0A. Genes Dev 19: 2236–2244.
13. Hernday AD, Braaten BA, Broitman-Maduro G, Engelberts P, Low DA (2004)Regulation of the pap epigenetic switch by CpxAR: phosphorylated CpxR
inhibits transition to the phase ON state by competition with Lrp. Mol Cell 16:537–547.
14. Hernday AD, Braaten BA, Low DA (2003) The mechanism by which DNA
adenine methylase and PapI activate the pap epigenetic switch. Mol Cell 12:
947–957.
15. Bigger J (1944) Treatment of Staphylococcal infections with penicillin byintermittent sterilisation. Lancet 244: 497–500.
16. Balaban NQ, Merrin J, Chait R, Kowalik L, Leibler S (2004) Bacterial
persistence as a phenotypic switch. Science 305: 1622–1625.
17. Kussell E, Leibler S (2005) Phenotypic diversity, population growth, andinformation in fluctuating environments. Science 309: 2075–2078.
18. Brauer MJ, Huttenhower C, Airoldi EM, Rosenstein R, Matese JC, et al. (2008)
Coordination of growth rate, cell cycle, stress response, and metabolic activity inyeast. Mol Biol Cell 19: 352–367.
19. Elliott B, Futcher B (1993) Stress resistance of yeast cells is largely independent ofcell cycle phase. Yeast 9: 33–42.
20. Lu C, Brauer MJ, Botstein D (2009) Slow growth induces heat-shock resistance
in normal and respiratory-deficient yeast. Mol Biol Cell 20: 891–903.
21. Gefen O, Balaban NQ (2009) The importance of being persistent: heterogeneity
of bacterial populations under antibiotic stress. FEMS Microbiol Rev 33:704–717.
22. Maisonneuve E, Shakespeare LJ, Jørgensen MG, Gerdes K (2011) Bacterial
persistence by RNA endonucleases. Proc Natl Acad Sci U S A 108:13206–13211.
23. Bar-Even A, Paulsson J, Maheshri N, Carmi M, O’Shea E, et al. (2006) Noise in
protein expression scales with natural protein abundance. Nat Genet 38:636–643.
24. Avery SV (2006) Microbial cell individuality and the underlying sources of
heterogeneity. Nat Rev Microbiol 4: 577–587.
25. Halme A, Bumgarner S, Styles C, Fink GR (2004) Genetic and epigenetic
regulation of the FLO gene family generates cell-surface variation in yeast. Cell116: 405–415.
26. Klar AJ, Srikantha T, Soll DR (2001) A histone deacetylation inhibitor and
mutant promote colony-type switching of the human pathogen Candidaalbicans. Genetics 158: 919–924.
27. Srikantha T, Tsai L, Daniels K, Klar AJ, Soll DR (2001) The histone deacetylase
genes HDA1 and RPD3 play distinct roles in regulation of high-frequencyphenotypic switching in Candida albicans. J Bacteriol 183: 4614–4625.
28. Becskei A, Kaufmann BB, van Oudenaarden A (2005) Contributions of low
molecule number and chromosomal positioning to stochastic gene expression.
Nat Genet 37: 937–944.
29. Xu EY, Zawadzki KA, Broach JR (2006) Single-cell observations revealintermediate transcriptional silencing states. Mol Cell 23: 219–229.
30. Levy SF, Siegal ML (2008) Network hubs buffer environmental variation in
Saccharomyces cerevisiae. PLoS Biol 6: e264. doi:10.1371/journal.pbio.0060264.
31. Tabor JJ, Bayer TS, Simpson ZB, Levy M, Ellington AD (2008) Engineering
stochasticity in gene expression. Mol Biosyst 4: 754–761.
32. das Neves RP, Jones NS, Andreu L, Gupta R, Enver T, et al. (2010) Connectingvariability in global transcription rate to mitochondrial variability. PLoS Biol 8:
e1000560. doi:10.1371/journal.pbio.1000560.
33. Colman-Lerner A, Gordon A, Serra E, Chin T, Resnekov O, et al. (2005)
Regulated cell-to-cell variation in a cell-fate decision system. Nature 437:699–706.
34. Volfson D, Marciniak J, Blake WJ, Ostroff N, Tsimring LS, et al. (2006) Origins
of extrinsic variability in eukaryotic gene expression. Nature 439: 861–864.
35. Vallen EA, Scherson TY, Roberts T, van Zee K, Rose MD (1992) Asymmetricmitotic segregation of the yeast spindle pole body. Cell 69: 505–515.
36. Aguilaniu H, Gustafsson L, Rigoulet M, Nystrom T (2003) Asymmetric
inheritance of oxidatively damaged proteins during cytokinesis. Science 299:1751–1753.
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 15 May 2012 | Volume 10 | Issue 5 | e1001325
37. Liu B, Larsson L, Caballero A, Hao X, Oling D, et al. (2010) The polarisome is
required for segregation and retrograde transport of protein aggregates. Cell140: 257–267.
38. Henderson KA, Gottschling DE (2008) A mother’s sacrifice: what is she keeping
for herself? Curr Opin Cell Biol 20: 723–728.39. Colman-Lerner A, Chin TE, Brent R (2001) Yeast Cbk1 and Mob2 activate
daughter-specific genetic programs to induce asymmetric cell fates. Cell 107:739–750.
40. Kale SP, Jazwinski SM (1996) Differential response to UV stress and DNA
damage during the yeast replicative life span. Dev Genet 18: 154–160.41. Slutsky B, Buffo J, Soll DR (1985) High-frequency switching of colony
morphology in Candida albicans. Science 230: 666–669.42. Srikantha T, Zhao R, Daniels K, Radke J, Soll DR (2005) Phenotypic switching
in Candida glabrata accompanied by changes in expression of genes withdeduced functions in copper detoxification and stress. Eukaryot Cell 4:
1434–1445.
43. Brockert PJ, Lachke SA, Srikantha T, Pujol C, Galask R, et al. (2003)Phenotypic switching and mating type switching of Candida glabrata at sites of
colonization. Infect Immun 71: 7109–7118.44. Levin-Reisman I, Gefen O, Fridman O, Ronin I, Shwa D, et al. (2010)
Automated imaging with ScanLag reveals previously undetectable bacterial
growth phenotypes. Nat Methods 7: 737–739.45. Ratcliff WC, Denison RF (2010) Individual-level bet hedging in the bacterium
Sinorhizobium meliloti. Curr Biol 20: 1740–1744.46. Thattai M, van Oudenaarden A (2004) Stochastic gene expression in fluctuating
environments. Genetics 167: 523–530.47. Acar M, Mettetal JT, van Oudenaarden A (2008) Stochastic switching as a
survival strategy in fluctuating environments. Nat Genet 40: 471–475.
48. Fraser D, Kaern M (2009) A chance at survival: gene expression noise andphenotypic diversification strategies. Mol Microbiol 71: 1333–1340.
49. King OD, Masel J (2007) The evolution of bet-hedging adaptations to rarescenarios. Theor Popul Biol 72: 560–575.
50. Contamine V, Picard M (2000) Maintenance and integrity of the mitochondrial
genome: a plethora of nuclear genes in the budding yeast. Microbiol Mol BiolRev 64: 281–315.
51. Dunn CD, Lee MS, Spencer FA, Jensen RE (2006) A genomewide screen forpetite-negative yeast strains yields a new subunit of the i-AAA protease complex.
Mol Biol Cell 17: 213–226.52. Lynch M, Sung W, Morris K, Coffey N, Landry CR, et al. (2008) A genome-
wide view of the spectrum of spontaneous mutations in yeast. Proc Natl Acad
Sci U S A 105: 9272–9277.53. Drake JW (1991) A constant rate of spontaneous mutation in DNA-based
microbes. Proc Natl Acad Sci U S A 88: 7160–7164.54. Hall DW, Mahmoudizad R, Hurd AW, Joseph SB (2008) Spontaneous
mutations in diploid Saccharomyces cerevisiae: another thousand cell genera-
tions. Genet Res (Camb) 90: 229–241.55. Orr HA (2009) Fitness and its role in evolutionary genetics. Nat Rev Genet 10:
531–539.56. Tirosh I, Reikhav S, Sigal N, Assia Y, Barkai N (2010) Chromatin regulators as
capacitors of interspecies variations in gene expression. Mol Syst Biol 6: 435.57. Vuorio OE, Kalkkinen N, Londesborough J (1993) Cloning of two related genes
encoding the 56-kDa and 123-kDa subunits of trehalose synthase from the yeast
Saccharomyces cerevisiae. Eur J Biochem 216: 849–861.58. Winderickx J, de Winde JH, Crauwels M, Hino A, Hohmann S, et al. (1996)
Regulation of genes encoding subunits of the trehalose synthase complex inSaccharomyces cerevisiae: novel variations of STRE-mediated transcription
control? Mol Gen Genet 252: 470–482.
59. Crowe JH, Hoekstra FA, Crowe LM (1992) Anhydrobiosis. Annu Rev Physiol54: 579–599.
60. Singer MA, Lindquist S (1998) Multiple effects of trehalose on protein folding invitro and in vivo. Mol Cell 1: 639–648.
61. Bandara A, Fraser S, Chambers PJ, Stanley GA (2009) Trehalose promotes the
survival of Saccharomyces cerevisiae during lethal ethanol stress, but does notinfluence growth under sublethal ethanol stress. FEMS Yeast Res 9: 1208–1216.
62. Hottiger T, Boller T, Wiemken A (1987) Rapid changes of heat and desiccationtolerance correlated with changes of trehalose content in Saccharomyces
cerevisiae cells subjected to temperature shifts. FEBS Lett 220: 113–115.63. Kandror O, Bretschneider N, Kreydin E, Cavalieri D, Goldberg AL (2004)
Yeast adapt to near-freezing temperatures by STRE/Msn2,4-dependent
induction of trehalose synthesis and certain molecular chaperones. Mol Cell13: 771–781.
64. Huh W-K, Falvo JV, Gerke LC, Carroll AS, Howson RW, et al. (2003) Globalanalysis of protein localization in budding yeast. Nature 425: 686–691.
65. Di Talia S, Skotheim JM, Bean JM, Siggia ED, Cross FR (2007) The effects ofmolecular noise and size control on variability in the budding yeast cell cycle.
Nature 448: 947–951.
66. Ghaemmaghami S, Huh W-K, Bower K, Howson RW, Belle A, et al. (2003)
Global analysis of protein expression in yeast. Nature 425: 737–741.
67. Chung JD, Stephanopoulos G, Ireton K, Grossman AD (1994) Gene expression
in single cells of Bacillus subtilis: evidence that a threshold mechanism controlsthe initiation of sporulation. J Bacteriol 176: 1977–1984.
68. Kearns DB, Losick R (2005) Cell population heterogeneity during growth ofBacillus subtilis. Genes Dev 19: 3083–3094.
69. Bitterman KJ, Medvedik O, Sinclair DA (2003) Longevity regulation inSaccharomyces cerevisiae: linking metabolism, genome stability, and hetero-
chromatin. Microbiol Mol Biol Rev 67: 376–399, table of contents.
70. Mortimer R, Johnston JR (1959) Life span of individual yeast cells. Nature 183:
1751–1752.
71. Egilmez NK, Jazwinski SM (1989) Evidence for the involvement of a
cytoplasmic factor in the aging of the yeast Saccharomyces cerevisiae.J Bacteriol 171: 37–42.
72. Kurec MM, Baszczynski M, Lehnert R, Mota A, Teixeira JA, et al. (2009) Flow
cytometry for age assessment of a yeast population and its application in beer
fermentations. J Inst Brew 115: 253–258.
73. Lesur I, Campbell JL (2004) The transcriptome of prematurely aging yeast cells
is similar to that of telomerase-deficient cells. Mol Biol Cell 15: 1297–1312.
74. Macarthur BD, Ma’ayan A, Lemischka IR (2009) Systems biology of stem cell
fate and cellular reprogramming. Nat Rev Mol Cell Biol 10: 672–681.
75. Huang S, Ingber DE (2006) A non-genetic basis for cancer progression andmetastasis: self-organizing attractors in cell regulatory networks. Breast Dis 26:
27–54.
76. Kaneko K (2011) Characterization of stem cells and cancer cells on the basis of
gene expression profile stability, plasticity, and robustness. Bioessays 33:
403–413.
77. Gupta PB, Fillmore CM, Jiang G, Shapira SD, Tao K, et al. (2011) Stochasticstate transitions give rise to phenotypic equilibrium in populations of cancer
cells. Cell 146: 633–644.
78. Roesch A, Fukunaga-Kalabis M, Schmidt EC, Zabierowski SE, Brafford PA,
et al. (2010) A temporarily distinct subpopulation of slow-cycling melanoma cells
is required for continuous tumor growth. Cell 141: 583–594.
79. Kitano H (2004) Biological robustness. Nat Rev Genet 5: 826–837.
80. Huang S (2009) Non-genetic heterogeneity of cells in development: more than
just noise. Development 136: 3853–3862.
81. Zhuravel D, Fraser D, St-Pierre S, Tepliakova L, Pang WL, et al. (2010)
Phenotypic impact of regulatory noise in cellular stress-response pathways. SystSynth Biol 4: 105–116.
82. LaFleur MD, Kumamoto CA, Lewis K (2006) Candida albicans biofilmsproduce antifungal-tolerant persister cells. Antimicrob Agents Chemother 50:
3839–3846.
83. McCusker JH, Clemons KV, Stevens DA, Davis RW (1994) Genetic
characterization of pathogenic Saccharomyces cerevisiae isolates. Genetics136: 1261–1269.
84. Pan X, Yuan DS, Xiang D, Wang X, Sookhai-Mahadeo S, et al. (2004) A robusttoolkit for functional profiling of the yeast genome. Mol Cell 16: 487–496.
85. Goldstein A, McCusker J (1999) Three new dominant drug resistance cassettesfor gene disruption in Saccharomyces cerevisiae. Yeast 15: 1541–1553.
86. Gietz RD, Woods RA (2002) Transformation of yeast by lithium acetate/single-stranded carrier DNA/polyethylene glycol method. Meth Enzymol 350: 87–96.
87. Seger J, Brockmann H (1987) What is bet-hedging? Oxford surveys inevolutionary biology, vol. 4. Oxford: Oxford University Press. pp 182–211.
88. de Jong IG, Haccou P, Kuipers OP (2011) Bet hedging or not? A guide to proper
classification of microbial survival strategies. Bioessays 33: 215–223.
89. Beaumont HJE, Gallie J, Kost C, Ferguson GC, Rainey PB (2009) Experimental
evolution of bet hedging. Nature 462: 90–93.
90. Acar M, Mettetal JT, van Oudenaarden A (2008) Stochastic switching as a
survival strategy in fluctuating environments. Nat Genet 40: 471–475.
91. Simons AM (2009) Fluctuating natural selection accounts for the evolution of
diversification bet hedging. Proc Biol Sci 276: 1987–1992.
92. Simons AM (2011) Modes of response to environmental change and the elusive
empirical evidence for bet hedging. Proc Biol Sci 278: 1601–1609.
93. Clauss MJ, Venable DL (2000) Seed germination in desert annuals: an empirical
test of adaptive bet hedging. Am Nat 155: 168–186.
94. Donaldson-Matasci MC, Lachmann M, Bergstrom CT (2008) Phenotypic
diversity as an adaptation to environmental uncertainty. Evol Ecol Res 10:493–515.
Bet Hedging in Yeast
PLoS Biology | www.plosbiology.org 16 May 2012 | Volume 10 | Issue 5 | e1001325